What Is Membership Inference?
Membership Inference is an attack that determines whether a specific data record was part of a model's training dataset by analysing the model's output behaviour for that record.
Membership Inference — an attack that determines whether a specific data record was part of a model's training dataset by analysing the model's output behaviour for that record.
Membership inference is a privacy attack with significant regulatory implications. If an attacker can determine that a specific individual's data was used to train a model, this may constitute a privacy breach — especially where the individual did not consent to their data being used for AI training. GDPR Article 17 (right to erasure) creates obligations that membership inference makes hard to satisfy: deleting a record from a database may not remove its influence from a trained model. Machine unlearning research addresses this problem.
Source: Shokri et al. (2017); GDPR, Article 17
Plain-language explanation
Membership inference is a privacy attack with significant regulatory implications. If an attacker can determine that a specific individual's data was used to train a model, this may constitute a privacy breach — especially where the individual did not consent to their data being used for AI training. GDPR Article 17 (right to erasure) creates obligations that membership inference makes hard to satisfy: deleting a record from a database may not remove its influence from a trained model. Machine unlearning research addresses this problem.
See where you stand on AI governance
Take the free 7-question maturity assessment and get a personalised action plan.
Free assessment — 3 minutes →